Thanks for the clarification. 2.8 isn’t stable enough for my purposes at the moment.
Cheers,
Ian
But you can use 2.79 master build with hybrid renderig. The are not as stable as 2.79 official releases but far more stable that 2.8. Just go to https://builder.blender.org/download/ and download 2.79 from there.
1 Like
Indeed, I use 2.79.4 which is pretty much fine but gives me hybrid rendering.
Same me. But undo/redo system is unstable and gives me crushes very often.
I have had a couple of crashes, but wasn’t really paying attention as to the cause. Shall keep my eye open for that in future.
As it seems the hybrid rendering is not working with osl scripts. Is it really so? Did anyone experimented with such a situation?
You mean Open Shading Language? As far as I know, OSL never worked in GPU (You do not have OSL option available if you choose GPU compute).
Yes Yafu, I do mean the Open Shading Language. I did not know that it was not working with GPU. Thanks for the information!
Can you guys help me
I have a system with windows 10, a gtx 1080 and ryzen 1600
And while in older builds CPU+gpu rendering working very well,
In the latest builds this combination it’s give me slower resaults
Are you guys in the same boat?
What tiles size? First builds with hybrid were fast on 16x16 but with lates you should use 32 and bigger. Just find the best ones for the scene and your rig.
I have try every combination of tile
16x16,32x32,64x64,even 480x270 but nothing the CPU+gpu setup it’s give me always lower resaults 
Ok here is the update!
In heavy scenes I see big difference but in a less heavy I see worst resaults
In simple scenes, I have found little or no variation and yes, even on the odd occasion a slight slowdown. However, nothing major, we are talking about 1 or 2 seconds difference at most. Going the other way to more complex scenes and the hybrid CPU+GPU really shines.
3 Likes
Just out of interest here’s some of my results. On a very recent 2.8 alpha build.
Ryzen 1700x + GTX 1070
complex scene with SSS, Volumetrics,complex shaders, large textures, large HDRI, light amount of triangles 3.8 million. 150spp.
16x16 tiles CUDA +CPU = 5:47 
32x32 tiles CUDA + CPU = 5:27 
Just CPU 32x32 tiles = 23:17 
Have you tried CUDA GPU only, without CPU?
And did you use CPU Cycles mode or GPU mode with your CPU as only enabled CUDA device?
CPU is plain old CPU mode, I selected Device: CPU in the scene options. As for GPU only mode im not sure if it has that or how to enable it… When i select Device =GPU task manager clearly shows that CPU+GPU are working together.
You can find it here, in System tab of User Preferences. Would be interesting to see if you turn of CPU, how much it actually contributes to the speed of GPU rendering. Often, in case of CPU+GPU rendering, CPU doesn’t tend to contribute that much. In fact, in very simple scenes, adding CPU as another CUDA device to very powerful GPU can slow the rendering down a little.
I disabled the cpu as a cuda device, and now rendered gpu only, as theres only one render tile doing the work.
CUDA only 32x32 tiles = 7:00
You should try 64 tiles for GPU+CPu - I get the best result on this tile size, and for only GPU - at least 96.
cool! i shall try.